1.48k likes | 1.65k Views
Temporal Information Systems in Medicine. Elpida Keravnou-Papailiou Summer School on Intelligent Systems 2 July 2007. Overview. The role of time in medicine Temporal modeling and temporal reasoning Temporal clinical databases Abstraction of time-oriented clinical data
E N D
Temporal Information Systems in Medicine Elpida Keravnou-Papailiou Summer School on Intelligent Systems 2 July 2007
Overview • The role of time in medicine • Temporal modeling and temporal reasoning • Temporal clinical databases • Abstraction of time-oriented clinical data • Time in clinical diagnosis • Research challenges
1. The role of time in medicine Science or Art? Knowledge-intensive to data-intensive applications systems that advice to systems that inform
Medical Tasks • Diagnose the cause of a problem • Predict its development • Prescribe treatment • Monitor the progress of a patient • Manage a patient computer-based support
Change in Focus The major challenge is no longer the deployment of knowledge but the intelligent exploitation of data
Exploitation of medical data • Extremely valuable and multifaceted • Can yield significant new knowledge • Can provide accurate predictors for critical risk groups based on “low-cost” information • Enables the intelligent comprehension of individual patients’ data • Closes the gap (conceptual distance) between raw patient data and medical knowledge
The change in focus has not changed the ultimate objective that still is … … to aid care providers reach the best possible decisions for any patient, to help them see through the consequences of their decisions/actions and if necessary to take rectifying actions as timely as possible
The change in focus has given rise to …. • Methods of abstraction, query and display of time-oriented data of relevance to all medical tasks, and • Has given a new dimension of significance to clinical databases, particularly the intelligent management and comprehension of clinical data
Time plays a major role in Medical Information Systems • Events occur at some time points • Certain facts hold during a time period • Temporal relationships exist between facts and/or events Abstracting time away means that dynamic situations are converted to static (snap-shot) situations, where neither the evolution of disorders, nor patient states can be modeled
Temporal Information Systems in Medicine are … • Information systems able to store, manage and query time-oriented clinical data, and • Support different inference tasks on these data
Research Directions • Temporal data maintenance – storage and retrieval of data with heterogeneous temporal dimensions • (temporal) database community • Temporal data abstraction and reasoning – supports various inference tasks involving time-oriented data • artificial intelligence community • Design of medical temporal systems • medical informatics community
2. Temporal modeling and temporal reasoning • Modeling temporal concepts • modeling time • modeling temporal entities • Temporal reasoning • General theories of time and the medical domain • Temporal constraints
Modeling Temporal Concepts -Modeling Time • Modeling time as a dense or discrete number line may not provide the appropriate abstraction for medical applications. Several basic choices have to be made: • Time domain • Instants and intervals • Linear, branching and circular times • Relative and absolute times • Temporal relationships • Granularities • Indeterminacy
Time Domain • (T; ≤), where T is a non-empty set of instants and ≤ is a total order on T • bounded • Contains upper and/or lower bounds w.r.t. to order relationship • or unbounded? • dense • ti, tj T with ti < tj, tkT s.t. ti< tk< tj • or discrete? • Every element has both an immediate successor and an immediate predecessor
Instants and Intervals • Time points represent instantaneous events, e.g. myocardial infarction • Time intervals represent situations lasting for a span of time, e.g. drug therapy • Often time points are the basic entities, where time intervals are represented by their start and end time points • Nonconvex intervals are intervals formed from a union of convex intervals and might contain gaps – can be used to represent processes or tasks that occur repeatedly over time
Nonconvex interval Convex interval Time
Linear, Branching and Circular Times • In reality, time is linear: the set of time points is completely ordered • For tasks such as diagnosis, projection or forecasting, a branching time might be necessary: hypothesizing possible past or future events/evolutions • Circular (or periodic) time is needed to describe times related to recurrent events, e.g. “administration of regular insulin every morning”
Relative and Absolute Times • The position on the time axis of an interval or of an instant can be given • As an absolute position, e.g. “on July 2, 2007” • As a relative position, e.g. “the day after”, or “sometime before now” • Absolute times are generally associated to a metric, being its position given as a distance from a given time origin, e.g. time origin is “admission to hospital”, metric is “days”, and absolute times could be “-3”, “4”, etc. • When a metric is defined for the time domain, relative times can be given quantitatively, e.g. “three days after birth”
Temporal Relationships • Allen’s interval algebra has been widely used in medical informatics • Various extensions have been proposed • Two main types of temporal relationships • Qualitative: interval I1before interval I2 • Quantitative: interval I1two hours before interval I2 • Can be classified according to the entities involved • Interval/interval, interval/point, point/interval, point/point
Allen’s 13 possible relations between time intervals A A A is EQUAL to B B is EQUAL to A A STARTS B B is STARTED BY A B B A A A FINISHES B B is FINISHED BY A A is BEFORE to B B is AFTER to A B B A A A is DURING B B CONTAINS A A MEETS B B is MET BY A B B A A OVERLAPS B B is OVERLAPPED BY A B
Granularities • Granularity: level of abstraction at which temporal information is expressed • Different units of measure allow different granularities • Clifford’s proposal • Assumes a chronon for every temporal domain • Constructive intervallic partitioning: segment the time line into mutually exclusive and exhaustive intervals • Temporal universe: hierarchy of time levels and units with defined semantics
Granularities • Bettini et al framework: a granularity is a mapping G from the integers (the index set) to subsets of the time domain s.t. • If i < j and G(i) and G(j) are non-empty, then all elements of G(i) are less than all elements of G(j), and • If i < k < j and G(i) and G(j) are non-empty, then G(k) is non-empty • Any G(i) is called a granule – the set of granules is discrete, irrespective of whether the the time domain is discrete or dense
Granularities • Bettini et al framework (cont.) • Besides an index, a granule may have a textual representation, e.g. “July 2007” • Relationships between granularities, e.g. • A granularity G1 is finer than another granularity G2 if for each i, there exists j such that G1(i)⊆G2(j) • Time-axes framework: conceptual modeling of time periods on the universal time line, each with its appropriate time-unit (granularity), e.g. infancy, childhood, puberty, etc. • Hierarchical and other relationships between time-axes • Spanning time-axes
Indeterminacy • Often the knowledge of when the considered fact happened is incomplete • We might not know precisely when a proposition became true and when it ceased to be true, although we might know that it was true during a particular time interval • The problem may arise because the time units involved have different granularities, or due to the naturally incomplete information in clinical settings • There is a need to model such vagueness • variable interval I, is composed of three consecutive convex intervals: begin(I), body(I), end(I)
Modeling Temporal Concepts -Modeling Temporal Entities • A rich model providing a number of interrelated basic temporal entities, different abstraction levels and multiple granularities is often required • In general, there are two approaches: • Adding a temporal dimension to existing objects • Database research: simple “atomic” entities • Creating model-specific, time-oriented entities • AI research: complex, task-specific entities
Example • Events • Instant-based objects • E.g. patient visits • Therapies • Time-interval objects • E.g. administration of drugs • Phases (of therapy) • Complex objects including events (visits) and therapies • Objects connected through a network
Kahn and colleagues Temporal Network (TNET) concept • Later evolved to Extended TNET (ETNET) • A T-node (or an ET-node) models task-specific temporal data (e.g. chemotherapy cycle) at different levels of abstraction • It is associated with a time interval during which the information represented by the T-node’s data is true for a given patient • Other systems inspired by the TNET model (e.g. M-HTP: monitoring heart-transplant patients)
Associating entities to instants or intervals • Defining occurrences of temporal entities, i.e. associating time with them: • Use both instants and intervals • Use just intervals, dealing in a homogeneous way with intervals degenerating to instants • Use just instants (time points)
Actual occurrences of temporal entities can be specified in different ways • Absolute temporal occurrences • Relative to some fixed time point, by specifying its initiation and termination • “Tachycardia on November 3, 2005 from 6:30 to 6:45 pm” • A common approach in temporal databases • Relative temporal occurrences • By referring to other occurrences • “angina after a long walk”, “several episodes of headache during puberty” – qualitative relationships • “angina two hours before headache” – quantitative relationships
Occurrences of temporal entities • Absolute vagueness (indeterminacy) • Initiation and /or termination (and thus duration) cannot be precisely specified in a given temporal context – “precision” is relative to the particular temporal context • Earliest and/or latest possible time for initiation/ termination • Minimum and maximum for duration • “an atrial fibrillation episode occurred on Dec 14th, 2006 between 14:30 and 14:45 and lasted for three to four minutes”
Occurrences of temporal entities • Relative vagueness • An occurrence’s temporal relation with other occurrences is not precisely known but can only be expressed as a disjunction of primitive relations • “the patient had vomited before or during the diarrhea episode” • Incorporating disjunctions within a standard temporal database is still a difficult task • Incompleteness in the specification of occurrences is thus a common phenomenon
Compound occurrences • Types • Periodic • Repetition, in a regular or non-regular fashion • Temporal trend: describes a change, the direction of change, and the rate of change, e.g. “increasing blood pressure” • Temporal patterns, e.g. • A sequence of meeting trends • Periodic occurrences • A set of causally related occurrences • Multiple levels of abstraction, with two basic structural relations • Refinement: going down to component occurrences • Abstraction: from components going up to container occurrences
Contexts, causality and other temporal constraints • Context: represents a state of affairs, that when interpreted (logically) over a time interval, can change the meaning of one or more facts which hold within the context time interval • Causality • A central relation between occurrences • Changes are explained through causal relations, and time is intrinsically related to causality • Basic temporal principle: an effect cannot precede its cause • Temporal constraints • Causally unrelated occurrences, can also be temporally constrained, e.g. a periodic occurrence could be governed by the constraint “the repetition occurs every 4 hours”
Temporal Reasoning • The ability to reason about time and temporal relations is fundamental to almost any intelligent entity that needs to make decisions • The real world includes not only static descriptions, but also dynamic processes • Evolving situations • Taking actions Modeling change thus time
Time is inherently relevant when reasoning about the world • Natural-language processing • “by the time you get home, I would be gone for 3 hours” • Robotics • Temporal order of actions, length of time to perform actions • Causal reasoning • Temporal precedence/equivalence • Scheduling tasks in a production line • Serial and concurrent actions – time-intervals • Patterns in a baby’s psychomotor development • “walking typically starts when the baby is about 12 months old, and is preceded by standing”
Medical Tasks • Projection: hypothesizing the development of the patient’s state, e.g. after the application of some therapeutic action • Forecasting: predicting particular future values for various parameters given a vector of time-stamped past and present measured values • Planning: producing a sequence of actions for a care provider, given an initial state of the patient and a goal state • Actions are operators with certain or probabilistic effects on the environment • May require enabling preconditions
Medical Tasks • Interpretation: • abstraction of a set of time-oriented data, either to an intermediate level of meaningful temporal patterns (e.g. temporal-abstraction or monitoring) or to a level of definite explanation of the findings and symptoms (e.g. diagnosis) • Unlike projection and forecasting, interpretation involves reasoning about the present and not the future • General criterion for classifying temporal-reasoning research: deterministic or probabilistic?
Temporal Reasoning Requirements:Generic Functionalities • Mapping the existence of occurrences across temporal contexts • Determining bounds for absolute existences of occurrences • Consistency detection and clipping of uncertainty • Deriving new occurrences from other occurrences (temporal-abstraction, decomposition derivations, causal derivations, etc) • Deriving temporal relations between occurrences • Deriving the truth status of queried occurrences • Deriving the state of the world at a particular time
General Theories of Time and the Medical Domain • Allen’s time-interval algebra • Kowalski and Sergot’s Event Calculus • Dean and McDermott’s Time Map Manager
Allen’s time-interval algebra • Motivation: expression of natural-language sentences and representation of plans • Primitive: time-interval • Propositions interpreted over time-intervals • (instantaneous) events are degenerate intervals • 13 basic binary relations between time-intervals • Incomplete temporal information captured through disjunctions of temporal relations, e.g. • I1 < starts, finishes, during > I2
Allen’s time-interval algebra:Proposition Types • Properties hold over every subinterval of an interval, e.g. “John had fever during last night” • holds(p,t) ↔ ( t’ in(t’,t) → holds(p,t’)) • Events hold only over a whole interval and not over any subinterval of it, e.g. “John broke his leg on Saturday at 6 pm” • occur(e,t) & in(t’,t) → ~occur(e,t’) • Processes hold over some subintervals of the interval in which they occur, e.g. “John had atrial fibrillation during last month” • occurring(p,t) → t’ in(t’,t) & occurring(p,t’)
Allen’s time-interval algebra • No branching time, either in the past or the future • Two forms of causality: event and agentive • Transitivity table that defines the conjunction of any two relations • Rij & Rjk → Rik • sound but incomplete algorithm for propagating transitivity relations
Kowalski and Sergot’s Event Calculus (EC) • A theory of time and change, well founded and formally studied • Events happen at time points and initiate and/or terminate time intervals over which properties hold • initiates(ev1, prop, t) ← happens(ev1, t) & holds(prop1, t) & …. & holds(propN, t) • terminates(ev1, prop, t) ← happens(ev1, t) & holds(prop1, t) & …. & holds(propN, t)
Event Calculus (EC) • default persistence • Initiated properties are assumed to persist until the occurrence of an event that interrupts them • EC is concerned with deriving the maximal validity intervals (MVIs) over which properties hold • do not contain any interrupting event for the property • are not subsets of any other validity intervals for the property • mholds_for(p, [S,E])
Weakly and Strongly Initiating and Terminating Events wI wI sT sT 0 1 2 3 4 5 6 7 sI sI sT sT 0 1 2 3 4 5 6 7 wI wI wT wT 0 1 2 3 4 5 6 7 sI sI wT wT 0 1 2 3 4 5 6 7
Interpreting initiating and terminating events • Different choices may give different MVIs • The choice depends on the property to be modeled, e.g. • Weak initiates supports aggregation • In patient monitoring we need to aggregate similar observed situation, so that a transition in the classification of the patient situation (say ventilatory state) is not caused • Strong initiates supports omission • Useful when dealing with incomplete sequences of events, e.g ECG monitoring (connect, connect, disconnect)
McDermott’s temporal logic • Goal: to model causality and continuous change and to support planning • Primitive: time points, where time is dense (the time line is the set of real numbers) • States: instantaneous snapshots of the universe (order preserving function: date) • Intervals: ordered pairs of states
McDermott’s temporal logicProposition Types • Fact: (T s p) where s, a state, and p, a proposition • Interpreted over points: “p is True in s” • A proposition, e.g. (On Patient1 Bed2), represents the states where Patient1 is on Bed2 • Event, e, is the set of intervals over which the event exactly happens • (Occ s1 s2 e): event e occurred between the states s1 and s2, i.e. over the interval [s1 s2]
McDermott’s temporal logic chronicles • states are totally ordered for the past, but are partially ordered and branching into the future • This branching captures the notion of a known past, but an indefinite future • Chronicle • A maximal linear path in such a branching tree • Represents a complete possible history of the universe, extending to the indefinite past and future, i.e. a totally ordered set of states that extends infinitely in time
Dean and McDermott’s Time Map Manager (TMM) • Temporal primitive: time point (instant) • Time-token: an interval together with a (fact or event) type • Time map: a collection of time-tokens represented as a graph • Nodes denote instants of time associated with the beginning and ending of events • Arcs describe relations between pairs of instants • Can be applied under a dense or a discrete model of time